Introduction to mere.run plugins

This guide is for people who run local AI production workflows or automate those workflows. It explains the boundary between mere.run and its official companion plugins.

Understand what plugins add

Plugins turn individual inference commands into repeatable workflows. A plugin can:

• Validate tools, inputs, credentials, and output locations.
• Write an inspectable plan before execution.
• Call the installed mere.run runtime for canonical model behavior.
• Coordinate local tools or user-controlled providers.
• Record artifacts, hashes, status, and cleanup in a run.json file.
• Provide machine-readable output for agents and pipelines.

Plugins run as separate executables. mere.run doesn't load plugin code into the core process.

Understand what stays in the core runtime

mere.run owns model discovery, download, storage, loading, and inference. A plugin can select and compose core commands, but it must not ship a competing model runtime.

This boundary keeps local inference behavior consistent. It also isolates plugin dependencies, permissions, provider failures, and upgrades.

Choose a workflow category

The catalog contains 18 official commands across the following workflow categories:

Category	Example tasks
Create media	Generate images, prepare visual-effects shots, produce animatics, make short films, and perform music
Process data	Analyze geospatial sources, search photo libraries, index archives, convert documents, and transcribe audio
Automate work	Compile graphs, prepare datasets, run command batches, and preserve resumable manifests
Use providers	Train adapters on RunPod and publish review artifacts to Flow Production Tracking
Evaluate systems	Compare local agents with Terminal-Bench and run identity-adapter evaluations

For a task-to-command mapping, see Choose a plugin. For the complete catalog, see Official plugins.

Apply the operating guarantees

Official plugins follow these guarantees:

• Local workflows keep their source data on the local machine unless the documented workflow uses a provider.
• Provider resources remain in your account.
• Paid work appears in a plan before it starts.
• Remote resources terminate by default.
• Commands that promise JSON write machine-readable data to stdout.
• Commands write human-readable diagnostics to stderr.
• A durable run manifest connects inputs, execution, artifacts, and cleanup.

To install and run a plugin, continue to Getting started.